Quantopian's community platform is shutting down. Please read this post for more information and download your code.
Back to Community
Interesting idea: cross-sectional bootstrapping for factor models
4 responses

So does one have to do 1000 backtests per day (or week or whatever) over a period of years to generate the "Cumulative factor model return in %" or am I missing something?

Yeah, I suppose to calculate the strategy paths in a non-survivorship-biased backtester, you'd need to sample from the cross-section at every decision point. You don't need to wait years for the results though, you do it all in a simulation.

Some while ago I tested a few of my strategies on random portfolios of US mutual funds with histories going back to 1985. 30,000 current and defunct funds in the sample available for selection. Yes it is helpful to get you to appreciate the dangers of curve fitting by portfolio section. But such a method based on past data will by definition be unable to produce much predictive performance value - let's face it, no amount of "math-turbation" can currently achieve that. Had my database consisted of world mutual funds the results would have shown a more robust view. But even so.....

Oh dear.... I always sound so horribly sceptical..... Of course this procedure is very valuable. But I know from personal experience over so many years that however hard I fiddle.... The future always turns out somewhat differently!